More than 650 photographers challenge contract language that may expose their work to AI training
Freelance photographers working for The Wall Street Journal are organising against a revised contractor agreement that would reportedly treat commissioned photographs as work made for hire and permit sublicensing without an express restriction on AI-training use. The campaign, led by Your Visual Colleagues and supported by the National Press Photographers Association, has attracted more than 650 photographers. The Journal says the changes are intended to protect its archive and that photographers would receive a joint copyright assignment allowing them to continue licensing their work.
Why visual creators should care
The dispute shows why ownership and AI permission cannot be separated. A photographer may technically retain some ability to license an image while a publisher simultaneously holds enough contractual control to provide that image, its metadata or unpublished outtakes to an AI company. Sensitive documentary photographs create additional risks because training access could include people, locations or identifying details deliberately excluded from publication.
Practical takeaway
Read work-for-hire, assignment and sublicensing clauses together. Require the contract to state whether published images, raw files, outtakes and metadata may be supplied for model training, fine-tuning, retrieval systems or other machine-learning uses. Do not rely on an editor’s verbal promise when the written agreement grants broader rights.
What to watch next
Watch whether the Journal adds an enforceable AI-training exclusion, limits sublicensing, changes its treatment of unpublished files or reaches a standard agreement with the photographers and NPPA. The outcome may influence contributor contracts across newspapers, magazines and stock archives.
Sources
EU sets a July 27 sign-up deadline for its AI-image transparency code
The European Commission has published updated operational guidance for organisations signing the Code of Practice on Transparency of AI-Generated Content. Providers and professional deployers that want to appear on the initial signatory list must submit the form by July 27 at 18:00 CEST. Eligible organisations include developers of systems that generate synthetic images or video, professional users responsible for disclosing deepfakes, and providers of provenance, watermarking and detection technology.
Why visual creators should care
The guidance turns the EU transparency framework into an immediate organisational decision. Image-generation companies and businesses professionally publishing deepfakes or other covered synthetic visuals may sign the code and use its measures to demonstrate compliance. Organisations that do not sign remain bound by Article 50 and may need to provide regulators with more extensive evidence showing that their alternative marking and disclosure system is adequate.
Practical takeaway
Determine whether the organisation is acting as a provider, deployer or both. Document how generated or manipulated images are marked in machine-readable form, when a visible deepfake label is required, how labels are presented accessibly and who checks that provenance survives export and publication.
What to watch next
Watch the Commission’s initial signatory list, which major image and video platforms join, and whether the pending transitional rule gives older generation systems until December 2 to meet machine-readable marking duties. Professional disclosure obligations are still scheduled to apply from August 2.
Sources
AI alteration did not erase the court artist’s copyright or moral rights
Dutch court artist Petra Urban received damages after a regional branch of Geert Wilders’ Party for Freedom used her courtroom drawing without permission and altered it with AI to make the depicted defendants appear more aggressive. The work was then used in political social-media material. The amount of compensation was not disclosed.
Why visual creators should care
The case demonstrates that running a protected image through an AI tool does not make the original creator’s rights disappear. The alteration also affected the meaning and neutrality of journalistic work, bringing moral rights and professional reputation into the dispute alongside ordinary licensing.
Practical takeaway
Photographers, illustrators and editorial artists should preserve the original work and capture every altered publication in context. A complaint should identify not only copying but also distortion, false political association, reputational damage and removal of the work from its original editorial setting.
What to watch next
Watch whether publishers and political organisations introduce approval rules for AI alterations of licensed editorial imagery and whether similar disputes produce formal court judgments rather than private settlements.
Sources
Photography competitions are banning AI faster than they are building verification systems
A review by Lumethic examined the published rules of 87 photography competitions. It found that 50 competitions explicitly banned AI-generated imagery, while another nine excluded it through camera-capture definitions. Only 27 reserved the right to request raw files, and none of the reviewed rulebooks mentioned C2PA or Content Credentials.
Why visual creators should care
A ban without a stated verification process can harm both the competition and legitimate photographers. Generated work may pass through unchecked, while authentic but unusual photographs may attract suspicion without a consistent evidentiary procedure.
Practical takeaway
Photographers entering competitions should retain raw files, layered edits, export histories and the rules in force on the submission date. Competition organisers should state what evidence may be requested and how disputed images will be examined before making public accusations.
What to watch next
Watch whether competitions begin accepting Content Credentials as supporting evidence, whether raw-file requirements expand and whether organisers publish appeal procedures for disputed or disqualified images.
Sources
World Press Photo treats generative enlargement as adding information to a photograph
World Press Photo’s 2026 rules prohibit synthetic imagery, generative fill and AI-powered enlargement tools that introduce new detail. Limited AI-assisted functions such as denoising, automatic adjustments and object selection may be permitted when they do not add, remove or materially alter information captured by the camera. Final-stage entrants must provide files recorded by the camera.
Why visual creators should care
The rules reject the idea that every AI-powered editing function is equivalent. They focus instead on whether a tool invents visual information or changes the documentary record, giving photographers a more practical boundary than a blanket prohibition on software containing AI.
Practical takeaway
Documentary photographers should check what an editing function actually does rather than relying on the product label. Keep the camera original, avoid tools that fabricate detail and save a clear edit history showing that permitted adjustments did not change the recorded scene.
What to watch next
Watch whether other press and documentary competitions adopt the same distinction between analytical assistance and generative alteration, and how computational smartphone modes are treated as they become harder to separate from conventional capture.
Sources
Generative functions are becoming ordinary parts of photography and design software
Adobe’s June Creative Cloud update introduced or expanded functions including AI-assisted culling, reflection removal, offline generative object removal and Illustrator’s Concept to Vector feature. The vector tool can turn sketches or low-quality source assets into editable drafts and produce stylistic variations from an existing image.
Why visual creators should care
The professional question is increasingly not whether AI software was used, but which function was used and whether it introduced new material. A photographer using automated culling or reflection isolation is in a different position from a designer generating new visual variations from source work.
Practical takeaway
Studios should record AI use at feature level. Project notes should distinguish selection, masking, denoising and isolation from generative removal, extension and concept generation, particularly when client contracts, contests or editorial rules prohibit invented content.
What to watch next
Watch whether Adobe exposes clearer feature-level provenance in exported files and whether agencies establish separate approval requirements for analytical automation and generative alteration.
Sources
Synthetic sets and likeness licensing are moving into mainstream film development
Recent film-industry reporting describes major filmmakers and production companies experimenting with AI-assisted pre-production, editing, visual effects, synthetic environments and digital likenesses. Reported projects include generated sets and locations intended to reduce physical production requirements, alongside performers licensing their likenesses or voices for controlled synthetic use.
Why visual creators should care
AI adoption in filmmaking is no longer limited to experimental shorts. When generated environments, performances or post-production elements enter major productions, agreements must address which crew roles are displaced, whose source work shaped the output and whether a performer’s digital identity can be reused beyond the original film.
Practical takeaway
Film productions should create shot-level AI records identifying generated environments, reference material, model versions, performer permissions and responsible human contributors. Digital-likeness licences should define the specific production, duration, territory, promotional use and whether future derivatives are permitted.
What to watch next
Watch union enforcement, distributor disclosure requirements, screen-credit practices and whether insurers demand source and consent records before covering AI-assisted productions.
Sources
Accessible generative tools can refine deepfakes until detectors miss them
Researchers found that commercially available generative AI systems could be prompted to refine synthetic facial images in ways that reduced detection while preserving the represented identity. The work did not require a new manipulation technique or specialist model access; it used ordinary user-facing systems and benign refinement instructions.
Why visual creators should care
The study exposes a mismatch between laboratory detection benchmarks and the tools available to ordinary users. A creator, editor or platform cannot assume that a deepfake detector remains reliable after an image has been passed through another generative system for apparently minor visual improvement.
Practical takeaway
Do not use a detector result as the sole basis for accusing a photographer, rejecting an image or authenticating a likeness. Combine detector outputs with source files, provenance records, publication history, identity checks and direct confirmation from the creator.
What to watch next
Watch whether detector vendors begin testing against iterative commercial-model refinement and whether generation platforms restrict prompts that explicitly optimise images against authenticity or detection criteria.
Sources
